CN103310235A - Steganalysis method based on parameter identification and estimation - Google Patents

Abstract

The invention discloses a steganalysis method based on parameter identification and estimation. The steganalysis method comprises the following steps of: (1) establishing a steganography configuration scheme knowledge base, wherein each configuration scheme comprises a configuration parameter Fi of a steganalysis classifier, and an attribute parameter vector Ti of a training sample used for acquiring the configuration parameter Fi; (2) as for a to-be-detected object, firstly identifying an attribute parameter vector P of the to-be-detected object; (3) carrying out similarity calculation on the attribute parameter vector P of the to-be-detected object and the attribute parameter vector Ti of each configuration scheme in the steganography configuration scheme knowledge base, so as to obtain a similarity measure index; and (4) selecting parameter configuration of the steganalysis classifier corresponding to the configuration scheme with the maximum similarity measure index as relevant parameters in the steganalysis classifier, carrying out steganalysis on the to-be-detected object, and judging whether the to-be-detected object is a steganography text sample containing steganography information or not. Compared with the prior art, the steganalysis method based on the parameter identification and estimation has the advantage that the accuracy rate of the steganalysis is greatly improved.

Description

A kind of based on parameter identification and the steganalysis method of estimating

Technical field

The present invention relates to a kind of steganalysis (Steganalysis) method, be specifically related to a kind of steganalysis method based on parameter identification and estimation, the method belongs to the sub-field of Information hiding in the field of information security technology.

Background technology

Along with the development of computer and network technologies, the use of digital multimedia is more and more general, and the modern steganography take digital multimedia as carrier has obtained to pay close attention to more and more widely.Hidden write can be under the prerequisite that does not affect carrier data perceived content and quality coil insertion device confidential information therein, true by the existence of hidden secret communication or kept secure, improved the safety of confidential data transmission or storage.It is reported, hidden writing utilized by lawless person and malicious code.Write correspondingly with hidden, steganalysis then is judge for analyzing whether data contain the technology of hidden information.Because the data redundancy of digital multimedia is larger, the modern hidden key character of writing is that carrier data mostly is multi-medium data; The general title is not original text through the hidden media of writing, and the media after hidden the writing are hidden literary composition.Although the hidden telescopiny of writing is difficult to be identified by human eye on the perceived content of carrier data and the impact of quality, its change to carrier data generally can be by the reacting condition of some statistical nature out.Steganalysis utilizes these that hidden telescopiny of writing is analyzed media data than more sensitive statistical nature, thus the existence of identification hidden information.

The hidden system of writing generally is made of steganographic algorithm and correlation parameter configuration, its input is original text (collection) and classified information (generally being the data after encrypting), output is the hidden literary composition (collection) of implicit classified information, wherein the larger parameter configuration of image latent writing analyzing influence is comprised the factors such as picture size and embedding rate.Modern Steganalysis is judged the technology that whether contains secret information in the digital media data as a kind of, at the secure context important in inhibiting.Different according to the scope of application, steganalysis can be divided into following three classes: (1) special-purpose steganalysis (Specific Steganalysis), it is only applicable to identify the hidden literary composition (list of references: Fridrich J by after the special steganographic algorithm processing, Goljan M.Practical Steganalysis of Digital Images:State of the Art[J] //Security and Watermarking of Multimedia Contents, 2002,4675:1-13.); (2) general steganalysis (Universal Steganalysis), hidden literary composition (list of references: Shi Y after it is applicable to identify and is processed respectively by a plurality of or multiclass steganographic algorithm, Chen C, Chen W.A Markov process based approach to effective attacking JPEG steganography[C] //Proceedings of8th International Workshop on Information Hiding, Virginia, USA, Jul.10-12, Berlin:Springer-Verlag, 2007:249-264.); (3) multi-Class Steganalysis (Multi-class Steganalysis), can identify the steganographic algorithm that hidden literary grace uses (list of references:

Fridrich J.Merging Markov and DCT features for multi-class JPEG steganalysis[C] //Proceedings of the Society of Photo-optical Instrumentation Engineers, Bellingham:SPIE Press, 2008:1-13.).Existing steganalysis is a kind of process of pattern classification, the recognition result of above special-purpose steganalysis and general steganalysis is " being hidden literary composition " or " not being hidden literary composition ", therefore be two class categorizing systems, and multi-Class Steganalysis need to further identify the algorithm that the hidden person of writing adopts, and is the multicategory classification system therefore.Training set used in the pattern classification is larger on the accuracy rate impact of steganalysis.Training set in the steganalysis is comprised of through the steganalysis feature extraction former collected works and hidden collected works, needs to use the steganalysis system that trains to detect the multimedia file of identification after sample to be analyzed refers to.The steganalysis system uses training set by the model and parameter configuration that the sorter training study goes out, and treats analytic target and whether contains secret information and judge.If the parameters of training set (such as image size, embedding rate, JPEG quality factor etc.) is basic identical with the corresponding parameters of sample to be analyzed, result of determination is generally comparatively accurate, otherwise then inaccurate.But in actual applications, the steganalysis person can't learn the employed steganographic algorithm of the hidden person of writing and corresponding configuration when generating sample to be analyzed, can only take blind steganalysis, namely in the situation of the algorithm of not knowing the hidden person's of writing employing and configuration, carry out the training of steganalysis system, and use the steganalysis system that trains that sample to be tested is analyzed, still, can not guarantee that in analysis the parameters of training set is as far as possible identical with the corresponding parameters of sample to be analyzed.Two class general steganalysis and multi-Class Steganalysis all can be regarded the different implementation methods of blind steganalysis as, and all there is defects in they.

Existing researchist has carried out more deeply research to concrete sorting technique and the general steganalysis feature of blind steganalysis, yet, but do not improve for defects specially.Take the image latent writing analysis as example, common situation is that the researchist adopts the image collection (list of references: Wang P that has special parameter and attribute in experiment, Liu F, Wang G, et al.Multi-class steganalysis for Jpeg stego algorithms[C] //Image Processing, 2008.ICIP2008.15th IEEE International Conference on.IEEE, 2008:2076-2079.), as adopt specific dimensions, quality factor and embedding rate prepare training set, yet testing image hidden write processing configuration and image attributes is uncertain in reality; Another kind method be the image blend that comprises all parameters is trained (list of references:

Fridrich J.Multiclass detector of current steganographic method for JPEG format[J] .IEEE Transactions on Information and Security.2008.3 (4) .635-650.), although improved its applicability, Detection accuracy is unsatisfactory.Therefore, how the deficiency for above-mentioned blind steganalysis research improves the blind steganalysis Detection accuracy under the real scene, is the problem that the needs in steganalysis research and development field solve.

Through patent consulting, existing related application situation is as follows in field of the present invention:

1) number of patent application is that 200610113185.2 Chinese patent " system and method that is used for steganalysis " discloses and a kind ofly detects and the steganalysis system of structure detection based on statistics.The core detection module of the system of this patent is integrated six statistics detection modules and two structure detection modules can be used the steganography of main flow to take common graphics/audio as carrier, and the ciphertext part that carries of hiding carries out reliable detection.This patent by to the testing result of a plurality of detection modules with unite get and syncretizing mechanism carry out cascading judgement, steganography method to present main flow has good applicability, the defectives such as individual module narrow application range, average recall rate is low, false alarm rate is high have been reduced, realized that high-quality, high-level efficiency to variety carrier type, multiple hidden WriteMode detect, has simultaneously good extendability, be convenient to integrated new detection module, the detection performance of upgrade-system.This patented method has only been described a concrete steganalysis system, do not consider to provide a more blanket steganalysis method for designing, the formation of the method improvement sorter training set of also considering to adopt parameter identification and estimating, therefore, this patented claim is obviously different from basic purpose, thinking and the specific implementation of present patent application.

2) number of patent application is that 200610018494.1 Chinese patent " based on the intelligent image steganalysis system of three-tier architecture " discloses a kind of intelligent image steganalysis system based on three-tier architecture.The method of this patent adopts three-tier architecture, make up special steganalysis system for type under the steganographic algorithm, utilize expert system that existing special-purpose steganalysis method is built up model bank and rule base, and by man-machine interaction continuous renewal steganalysis rule base, and adopt the principal element feature extraction to combine with sample image storehouse classification based training, improve counting yield and accuracy.Although this patented method has provided the framework of the existing steganalysis method of a kind of better use, but it does not consider to adopt the formation of parameter identification and the method improvement sorter training set of estimating, therefore, this patented claim is obviously different from basic purpose, thinking and the specific implementation of present patent application.

Summary of the invention

For the foregoing problems that exists in the existing Steganalysis, the purpose of this invention is to provide a kind of based on parameter identification and the steganalysis method of estimating.The present invention is by the identification of computer intelligence ground and every parameter that affects the steganalysis accuracy of estimating multimedia file to be measured, from previously prepared knowledge base, select best steganalysis sorter configuration that it is analyzed, improve the Detection accuracy of blind steganalysis.

The basic skills flow process that the present invention provides is:

1) preparation comprises the knowledge base of a large amount of allocation plans, wherein each allocation plan every property parameters of comprising the configuration parameter of a steganalysis sorter and obtaining the used training sample of these parameters.Wherein, the configuration parameter of steganalysis sorter is the one group of parameter that is used for classification that obtains by training process, with support vector machine (SVM the most frequently used in the steganalysis, support vector machine) sorter is example, and the parameter configuration of its sorter comprises for determining the classify support vector at interface and corresponding weights thereof etc.The property parameters of training sample comprises the various factors that may affect the steganalysis effect, such as sample-size, textural characteristics, the compression quality factor, file layout etc.

2) actual carry out steganalysis before, for sample to be tested, identify in advance or estimate its property parameters, the similarity calculating method based on the parameter distance regretional analysis that proposes according to the present invention, the similarity measurement index of the property parameters of each allocation plan in the prepared knowledge base in calculating sample to be tested property parameters and the above step 1).This Measure Indexes is that the present invention proposes first, and for assessment of the similarity degree between sample to be tested and the allocation plan, desired value is larger, illustrates that sample to be tested and allocation plan similarity are higher.

3) according to above step 2) result of calculation, select the allocation plan of similarity measurement index maximum, relevant parameter with in the wherein corresponding classifier parameters config update steganalysis sorter carries out steganalysis to sample to be tested again, obtains final classification results.

Overall technological scheme of the present invention comprises the steps (all methods all can be finished according to user's configuration and operational order by computer program):

1) preparation comprises the knowledge base of various configurations scheme

Set (the X that the many groups of preparation original text sample forms ₁, X ₂..., X _s, every group of original text sample set X wherein _iIn sample have unified property parameters Certain property parameters that represents i group sample is such as sample-size, textural characteristics, the compression quality factor, file layout etc.Use respectively every group of original text sample set X _iThe hidden civilian sample set Y that preparation is corresponding _i, with X _iAnd Y _iRespectively as former collected works and hidden collected works, by obtaining the parameter configuration F of steganalysis sorter after the sorter training _i, and with F _iWith corresponding property parameters T ⁱCombination is preserved stand-by as an allocation plan.Obtain to comprise thus the knowledge base of a plurality of allocation plans.

2) in advance identification and the property parameters of estimating sample to be tested before steganalysis

Before being about to carry out steganalysis, need to identify in advance or estimate the property parameters P=(α of multimedia file to be measured ₁, α ₂..., α _n).Wherein, the base attribute parameter of multimedia file (such as parameters such as the size of multimedia file, JPEG quality factors) can directly be obtained from fileinfo; Texture complication can use existing any one texture classifying method to be divided into high, medium and low three classes.

3) the similarity measurement index of computation attribute parameter

Calculate successively in the property parameters P of sample to be tested and the knowledge base property parameters T in every kind of allocation plan ⁱSimilarity.The present invention proposes a kind of Measure Indexes of assessing property parameters similarity in sample to be tested property parameters and the allocation plan, on this basis, provide system of selection and the flow process of allocation plan, and finally adopt the steganalysis classifier parameters configuration F in the selected allocation plan _iConfiguration steganalysis sorter is realized the more accurately hidden media identification of writing.The computing method of similarity measurement index and flow process are:

A) parameter vector normalization

Unit between each dimension of property parameters vector is not identical, therefore at first needs parameter vector is carried out normalized.Be provided with m training set, the sample standard deviation in each training set has identical property parameters vector.Order

, i ∈ 1,2 ..., m} represents the property parameters vector of the corresponding m of this a m training set allocation plan, to each characteristic dimension of each vector

Ask expectation value μ={ μ ₁, μ ₂..., μ _j..., μ _nAnd standard deviation s={s ₁, s ₂..., s _j..., s _n, wherein to each j ∈ 1,2 ..., n} has

${\mathrm{\μ}}_j=\frac{{\mathrm{\Σ}}_{i=1}^m{\mathrm{\α}}_j^i}{m}---\left(1\right)$

$s_j=\sqrt{\frac{{\mathrm{\Σ}}_{i=1}^m{({\mathrm{\α}}_j^i-{\mathrm{\μ}}_j)}^2}{m-1}}---\left(2\right)$

Then, carry out normalization by following formula:

${\tilde{a}}_j^i=\frac{a_j^i-{\mathrm{\μ}}_j}{s_j}---\left(3\right)$

Wherein,

Be that the j dimensional feature of i allocation plan property parameters vector is through the parameter vector value after the normalized.After the normalization pre-service, no matter be the property parameters vector T of allocation plan ⁱ, or attribute parameter vector P to be analyzed, all have same dimension.If no special instructions, parameter vector used among the present invention is all pretreated through normalization, and is easy for explaining, and hereinafter still uses symbol

Parameter vector value after the expression normalization.

B) similarity measurement index definition

For the similarity measurement index that illustrates that the present invention proposes, at first define n dimension sample to be tested property parameters vector P and each allocation plan n dimension attribute parameter vector T ⁱRange formula be

$D(T^i,P)={\mathrm{\Σ}}_{k=1}^n{w}_k|{\mathrm{\α}}_k-{\mathrm{\α}}_k^i|---\left(4\right)$

Wherein, D (T ⁱ, the P) distance between the parameter vector of expression sample to be tested and i allocation plan, α _kWith

Represent respectively sample to be tested property parameters vector P and i allocation plan property parameters vector T ⁱThrough k parameter after the normalized, w _kRepresent k parameter calculate Weighted distance and in shared proportion.w _kValue be according to the factor of some parameter representatives the influence degree size of steganalysis accuracy rate to be determined, its value can be based on experience, the practical function principle between each factor is determined.The w that adopts among relevant the present invention _kThe example of value can be referring to the embodiment part.Distance metric shown in the formula (4) also can use other distance definitions to replace, such as the Euclidean distance under the d dimension space:

$D(T^i,P)={\mathrm{\Σ}}_{k=1}^n{w}_k{|{\left(a_k\right)}^d-{\left(a_k^i\right)}^d|}^{\frac{1}{d}}$

Above distance can reflect the similarity between sample to be tested property parameters vector and the allocation plan property parameters vector to a certain extent, but in the ordinary course of things, the more applicable similarity expression formula form of above direct definition that can not fit like a glove in practice, it generally is its a functional transformation form, therefore, be necessary to obtain this form with regression analysis technique.For this reason, based on D (T ⁱ, P) the similarity measurement index M (T between definition sample to be tested and the allocation plan property parameters vector ⁱ, P) be

$M(T^i,P)=f\left(D(T^i,P)\right)=f\left({\mathrm{\Σ}}_{k=1}^n{w}_k\right|{\mathrm{\α}}_k-{\mathrm{\α}}_k^i\left|\right)---\left(5\right)$

Wherein, the f function is certain function of distance between assessment sample to be tested property parameters vector and the allocation plan property parameters vector, and its concrete form awaits obtaining with regression analysis technique.Formula (5) is by using the f function to the parameter vector range formula of this paper definition, be the equal of a correcting process to the common distance of original definition, to assess better the similarity degree between allocation plan property parameters vector and the object properties parameter vector to be analyzed.M (T ⁱ, value P) is larger, and the property parameters of the corresponding originally training sample of property parameters and allocation plan institute of expression sample to be tested is more approaching, otherwise represents more to keep off.The concrete form of function f can be determined by micro-judgment or regression analysis in the formula (5).The concrete grammar of regretional analysis is, with D (T ⁱ, P) as independent variable, with M (T ⁱ, P) as dependent variable, with allocation plan T ⁱThe classification accuracy rate of corresponding training sample set cross validation uses least square regression to carry out regretional analysis as the predicted value of dependent variable, obtains the concrete form of function f.Wherein the concrete grammar of cross validation is, with training sample set by a certain percentage random division be simulated training collection and simulation test collection, use sorter to train and the judgement of classifying, record the classification accuracy rate of simulation test collection as the result of a cross validation; Cross validation and calculate average accuracy as final cross validation accuracy repeatedly.The concrete form example of function f can be referring to the embodiment part.

4) choose allocation plan and carry out steganalysis

After above-mentioned steps is finished, according to the result of calculation of similarity measurement index, select the allocation plan T with sample to be tested property parameters vector similarity maximum ^r, it can be expressed as

T ^r=max _i=1,2,…,sM(T ⁱ,P)(6)

Wherein, S represents that the allocation plan that comprises in the knowledge base is total.Use T ^rThe sorter configuration parameter configuration steganalysis sorter that comprises carries out steganalysis to sample to be tested, judges whether it is the hidden civilian sample that contains hidden information.

The present invention comprises the effect of correlative technology field:

1) identifies and estimates the accuracy rate that has improved steganalysis based on parameter.By multimedia file to be measured being carried out parameter identification and estimating, subsequently, in the allocation plan of knowledge base (each comprise the property parameters vector that one group of training sample has and the sorter configuration parameter that obtains based on this group training sample), select targeted specifically to comprise the scheme that approaches the most with file attribute parameter to be measured, choose classifier parameters configuration steganalysis sorter wherein, with this sorter object to be measured is detected.Approach as much as possible on attribute owing to having guaranteed so intelligently training sample and sample to be tested, can improve the accuracy of steganalysis classification.

2) utilize more optimally option and installment scheme of regression analysis technique.The present invention proposes a kind of similarity measurement index, this index is based on the method for regretional analysis, with the base attribute parameter as vector, assessment in the allocation plan property parameters and the similarity degree between the object properties parameter to be analyzed, thereby can automatically filter out more excellent allocation plan.The property parameters that the latter comprises and the property parameters of sample to be tested are the most approaching, and this has guaranteed to use the sorter of the configuration parameter configuration in this scheme to be suitable for analyzing object to be measured most, and then has improved the accuracy rate of steganalysis.

Description of drawings

Fig. 1 is the overview flow chart of the inventive method;

Fig. 2 is the process flow diagram that the present invention calculates the similarity measurement index.

Embodiment

Propose a kind of based on parameter identification and the steganalysis method of estimating among the present invention.Its main process comprises preparation allocation plan knowledge base, identification and estimates sample to be tested property parameters vector, calculates sample to be tested and each allocation plan and comprise classifier parameters corresponding to index of similarity, the selection index of similarity maximum configured scheme attribute parameter of property parameters vector and configure the steganalysis sorter, carry out steganalysis etc.By the present invention, the steganalysis person can be by identification and the property parameters of estimating multimedia file to be measured, calculate the measuring similarity index of every allocation plan property parameters in this property parameters and the knowledge base, select the allocation plan of desired value maximum, use sorter configuration parameter configuration steganalysis sorter corresponding to this scheme, sample to be tested is analyzed, thus the accuracy of raising steganalysis.

Below in conjunction with accompanying drawing and take the jpeg image steganalysis as example, the present invention is done more concrete description.

Fig. 1 has described overall flow of the present invention.At first, preparation allocation plan knowledge base.After determining the property parameters of one group of original text sample, select or prepare an original text sample set, until obtain many group original text sample sets, wherein the sample standard deviation in every group of original text sample set has unified property parameters vector.Prepare corresponding hidden civilian sample set by the original text sample set, each hidden civilian sample set also has unified property parameters vector, and these former collected works and hidden collected works are carried out the training of steganalysis sorter as training set, will be stored as an allocation plan by sorter configuration parameter and the corresponding sample set property parameters that training obtains.Secondly, for image to be detected, extract or estimate its property parameters vector, calculate the measuring similarity index that itself and each allocation plan comprise the property parameters vector.At last, determine and the allocation plan property parameters vector of sample to be tested property parameters vector similarity index maximum, the classifier parameters of selecting to comprise in the corresponding configuration scheme configures the steganalysis sorter, and sample to be tested is detected.The embodiment of above-mentioned steps is below described:

1) determines sets of attribute parameters

Selection affects obvious factor (such as image texture, size, quality factor etc.) to steganalysis, as the one dimension of property parameters vector, namely determines parameter vector (α ₁, α ₂... α _n).Wherein to the parameter alpha of each dimension ₁, α ₂... α _n, arranging its value has respectively η ₁, η ₂... η _nKind.For example picture size can select greatly (3000 * 2000), in (1500 * 1000), little (512 * 512) three kinds, quality factor can select 80,90 two kind, texture can be selected three kinds of high, medium and low textures.In actual applications, the value kind on each dimension is more, and construction knowledge base is just more complete, and in use of the present invention, the similarity of property parameters vector can be higher, and the final sorter configuration parameter that obtains will be more reasonable.

2) preparation allocation plan storehouse

According to step 1) determine that parameter vector prepares some groups of former collected works, the sample standard deviation in every group of original text sample set has unified property parameters vector.Use jpeg image steganography method (such as F5, MME, nsF5, JSteg etc.) typical or that pay close attention to, former collected works are carried out hidden writing embed the corresponding hidden collected works of acquisition.Every group of former collected works and hidden collected works are extracted respectively the steganalysis feature, use support vector machine to train as sorter and obtain steganalysis sorter configuration parameter, configuration parameter and the property parameters of corresponding sample set are stored as an allocation plan, wherein, application of formula (3) has been carried out normalized to the property parameters vector.Finally obtain a knowledge base that comprises some allocation plans.

3) identification and the property parameters of estimating sample to be tested

Identify or estimate the property parameters P=(α of multimedia file to be measured ₁, α ₂..., α _n).Wherein, the base attribute parameter of multimedia file (such as parameters such as the size of multimedia file, JPEG quality factors) can directly be obtained from fileinfo; Texture complication can use existing any one texture classifying method to be divided into high, medium and low three classes.Application of formula (3) is carried out normalized to the property parameters vector.

4) the similarity measurement index of computation attribute parameter vector

Calculate the distance D (T that this attribute parameter vector to be analyzed and all allocation plans comprise the property parameters vector according to formula (4) ⁱ, P).Weight w _kCan rule of thumb obtain, wherein size, quality factor, weights that texture is corresponding can be followed successively by 0.5379,0.1614,0.0035.At last, calculate similarity measurement desired value M (T by formula (5) ⁱ, P), wherein the concrete form of function f obtains according to regretional analysis, and instantiation is

f(x)=(1.0211x+0.2047)/(1.5358x ²+0.2031)。

5) the option and installment scheme is carried out steganalysis

Select similarity measurement desired value M (T by formula (6) ⁱ, P) maximum allocation plan T ^r, use T ^rThe classifier parameters configuration steganalysis sorter that comprises carries out steganalysis to sample to be tested again, judges whether it is the hidden civilian sample that contains hidden information.

Claims (11)

1. the steganalysis method based on parameter identification and estimation the steps include:

1) sets up a hidden allocation plan knowledge base of writing; Wherein, each allocation plan comprises the configuration parameter F of a steganalysis sorter _iAnd obtain described configuration parameter F _iThe property parameters vector T of used training sample ⁱ

2) treat detected object, at first determine its property parameters vector P;

3) will this attribute parameter vector P to be measured and the described hidden property parameters vector T of writing each allocation plan in the allocation plan knowledge base ⁱCarry out similarity and calculate, obtain a measuring similarity index;

4) select the corresponding steganalysis classifier parameters of allocation plan of similarity measurement index maximum to configure, as the relevant parameter in the steganalysis sorter, this object to be measured is carried out steganalysis, judge whether it is the hidden civilian sample that contains hidden write information.

2. the method for claim 1 is characterized in that described measuring similarity index is property parameters vector P and property parameters vector T ⁱDistance.

3. method as claimed in claim 2 is characterized in that at first respectively to the described property parameters vector T in each allocation plan ⁱCarry out normalized, this attribute parameter vector P to be measured is carried out normalized; Then calculate described measuring similarity index.

4. method as claimed in claim 3 is characterized in that the described hidden allocation plan knowledge base of writing comprises m allocation plan, the property parameters vector T in each allocation plan ⁱBe n-dimensional vector; Namely , i ∈ 1,2 ..., m},

Be i property parameters vector T ⁱIn j characteristic dimension.

5. method as claimed in claim 4 is characterized in that according to formula

Computation attribute parameter vector P and each property parameters vector T ⁱSimilarity, obtain described measuring similarity and refer to target value; Wherein, D (T ⁱ, P) the property parameters vector T of expression object to be measured and i allocation plan ⁱBetween distance, α _kK parameter after the expression property parameters vector P normalized,

Expression property parameters vector T ⁱThrough k parameter after the normalized, w _kRepresent k parameter calculate Weighted distance and in shared proportion.

6. method as claimed in claim 4 is characterized in that according to formula Computation attribute parameter vector P and each property parameters vector T ⁱSimilarity, the Euclidean distance that obtains under the d dimension space refers to target value as described measuring similarity; Wherein, D (T ⁱ, P) the property parameters vector T of expression object to be measured and i allocation plan ⁱBetween distance, α _kK parameter after the expression property parameters vector P normalized,

Expression property parameters vector T ⁱThrough k parameter after the normalized, w _kRepresent k parameter calculate Weighted distance and in shared proportion.

7. such as the arbitrary described method of claim 3～6, it is characterized in that described property parameters vector T ⁱThe method of carrying out normalized is:

71) to described property parameters vector T ⁱEach characteristic dimension Ask expectation value μ={ μ ₁, μ ₂..., μ _j..., μ _nAnd standard deviation s={s ₁, s ₂..., s _j..., s _n; Wherein, ${\mathrm{\μ}}_j=\frac{{\mathrm{\Σ}}_{i=1}^m{\mathrm{\α}}_j^i}{m},$ $s_j=\sqrt{\frac{{\mathrm{\Σ}}_{i=1}^m{({\mathrm{\α}}_j^i-{\mathrm{\μ}}_j)}^2}{m-1}};$

72) by formula

Calculate described property parameters vector T ⁱIn the normalized value of j dimensional feature.

8. such as claim 5 or 6 described methods, it is characterized in that utilizing formula $M(T^i,P)=f\left(D(T^i,P)\right)=f\left({\mathrm{\Σ}}_{k=1}^n{w}_k\right|{\mathrm{\α}}_k-{\mathrm{\α}}_k^i\left|\right);$ To D (T ⁱ, P) revise; Wherein, function f () is for to obtain with regression analysis.

9. method as claimed in claim 8 is characterized in that the acquiring method of described function f () is: with D (T ⁱ, P) as independent variable, with M (T ⁱ, P) as dependent variable, with allocation plan T ⁱThe classification accuracy rate of corresponding training sample set cross validation uses least square regression to carry out regretional analysis as the predicted value of dependent variable, obtains the concrete form of function f ().

10. method as claimed in claim 9 is characterized in that described property parameters vector T ⁱIn, the parameter of each dimension has multiple value.

11. method as claimed in claim 10 is characterized in that described property parameters vector T ⁱComprise image texture characteristic, size characteristic, quality factor feature; The proportion of size characteristic is 0.5379, and the weights proportion of quality factor feature is 0.1614, and the proportion of image texture characteristic is 0.0035.

Images